#Giving access credentials
1 messages · Page 1 of 1 (latest)
Don’t paste any real API keys/passwords into Discord (even in a “credentials” channel) — treat this as public forever.
For a VPS, the practical “best practice” split is:
1) API keys / OAuth tokens OpenClaw needs to run
Store them on the VPS, on disk, in OpenClaw’s state, not in chat:
- Use auth profiles (what onboarding /
openclaw models auth …creates) for model providers. - Use env vars for everything else (tokens for integrations/skills).
Docs for how env loading works (and where .env is read from):
https://docs.openclaw.ai/environment
FAQ section: https://docs.openclaw.ai/help/faq#env-vars-and-env-loading
Security hardening / permissions:
https://docs.openclaw.ai/gateway/security (and run openclaw security audit)
2) “Usernames & passwords I want to ask the bot for later”
Honestly: don’t do this. OpenClaw (or any LLM agent) can’t guarantee “never reveal outside” if it’s capable of reading the secret; prompt-injection or a misfire could leak it.
Instead:
- Keep human passwords in a real password manager (1Password/Bitwarden/etc).
- Prefer API tokens / app passwords / OAuth scoped to the minimum permissions needed.
- If the goal is “bot can log into websites”: prefer doing that via a local node/browser takeover so the password stays on your personal machine, not on the VPS.
3) “Without SSH’ing into the server”
You still need some secure path to deliver secrets once, but you can avoid interactive SSH by using tailnet access:
- Keep the Gateway loopback-only and access it via Tailscale Serve (recommended) or a VPN/tailnet.
- Then manage auth/config remotely from your laptop using OpenClaw’s remote mode (the CLI talks to the Gateway over WebSocket and can apply config / auth without you logging into a server shell).
Remote access doc: https://docs.openclaw.ai/gateway/remote